跳到主内容
智客 ZICQ

技能库 智客分类:其他 Screenshot OpenClaw

Screenshot

抓取,检查,并比较屏幕,窗口,区域,网页,模拟器的截图,以及CI与正确的工具运行,等待策略,取景口,输出格式. 当 (1) 要调试,QA,docs,bug报告,或视觉审查需要截图时使用; (2) 涉及桌面,浏览器,模拟器,或无头抓取; (3) 稳定的截图需要固定的取景口,安放,遮掩,或动画控制.

534 安装量 · 38 星标

官方网址:作者主页

技能介绍

先看中文介绍;官方 description 原文单独保留,不改写 SKILL.md。

做什么

抓取、检查和比较屏幕、窗口、区域、网页、模拟器和 CI 的截图和正确的工具、等待策略、视图和输出格式

何时用

当任务需要桌面应用程序、浏览器页面、模拟器、区域、窗口或全屏幕截图时使用,特别是用于调试、QA、文档、发布注释、错误报告、视觉审查或比较前/之后

代理如何加载

按 Agent Skills 渐进披露:启动时只加载 name 与 description(约 100 token);任务匹配后才读入整份 SKILL.md 正文;scripts/、references/、assets/ 仅在需要时再读。 本文件正文结构:When to Use、Tool Choice、Core Rules、1. Pick the capture path by artifact, not by habit、2. Stabilize the target before capturing、3. Freeze viewport, scale, zoom, and theme for reproducibility。

文件分析

文件分析:这是一份仅含 SKILL.md 的指令型技能,代理激活后整份正文进入上下文。

官方 description(原文)

Capture, inspect, and compare screenshots of screens, windows, regions, web pages, simulators, and CI runs with the right tool, wait strategy, viewport, and output format. Use when (1) you need screenshots for debugging, QA, docs, bug reports, or visual review; (2) desktop, browser, simulator, or headless capture is involved; (3) stable screenshots require fixed viewport, settling, masking, or animation control.

When to UseTool ChoiceCore Rules1. Pick the capture path by artifact, not by habit2. Stabilize the target before capturing3. Freeze viewport, scale, zoom, and theme for reproducibility4. Capture the smallest useful scope5. Remove noise before you capture6. Use the right output format7. Make automation and CI captures debuggable8. Validate that the screenshot is actually usefulHigh-Value Patterns

来源分类:ClawHub Debugging

SKILL.md 与 Agent 调用

官方规范 ↗
name
Screenshot
description
Capture, inspect, and compare screenshots of screens, windows, regions, web pages, simulators, and CI runs with the right tool, wait strategy, viewport, and output format. Use when (1) you need screenshots for debugging, QA, docs, bug reports, or visual review; (2) desktop, browser, simulator, or headless capture is involved; (3) stable screenshots require fixed viewport, settling, masking, or animation control.
  1. 发现技能客户端向 Agent 提供名称与描述目录。
  2. 匹配与调用用户指定或任务匹配后,载入 SKILL.md 指令。
  3. 按需加载按步骤读取参考文档、使用脚本与素材。

具体调用语法与可用工具以目标 Agent 客户端为准。 查看调用机制说明 ↗

安装这个技能

Skills CLI ↗

先选择目标 Agent 和安装范围,保留技能包的附属文件,安装后检查客户端能否发现该技能。

交给 Agent 安装

复制安装指令给支持 Agent Skills 的代理,确认其中的目标目录与客户端匹配。

把 Agent Skill「Screenshot」安装到我的项目:SKILL.md 原文与官方 description 见 https://zicq.com/zh/skills/skl-ab82b8b86c9ab791-%E6%88%AA%E5%9B%BE.html
请存为 .cursor/skills/screenshot/SKILL.md 或 .claude/skills/screenshot/SKILL.md,frontmatter 的 name 与 description 保持原样,不要改写。

当前没有明确的 GitHub 技能包地址,请按来源页面的安装器说明操作。

ClawHub ↗

阅读排版
--- name: Screenshot slug: screenshot version: 1.0.1 homepage: https://clawic.com/skills/screenshot description: "Capture, inspect, and compare screenshots of screens, windows, regions, web pages, simulators, and CI runs with the right tool, wait strategy, viewport, and output format. Use when (1) you need screenshots for debugging, QA, docs, bug reports, or visual review; (2) desktop, browser, simulator, or headless capture is involved; (3) stable screenshots require fixed viewport, settling, masking, or animation control." changelog: "Improved screenshot guidance with stronger browser, simulator, CI, and visual-stability rules while keeping the skill compact." metadata: {"clawdbot":{"emoji":"📸","os":["linux","darwin","win32"]}} --- ## When to Use Use when the task needs a screenshot of a desktop app, browser page, simulator, region, window, or full screen, especially for debugging, QA, documentation, release notes, bug reports, visual review, or before/after comparison. This skill is about taking the right screenshot reliably, not about editing images after the fact. ## Tool Choice | Context | Best default | Why | |---------|--------------|-----| | macOS desktop or window | `screencapture` | Built-in, reliable, supports silent, interactive, region, and window capture | | iOS Simulator | `xcrun simctl io booted screenshot` | More reliable than generic desktop capture for simulator output | | Linux Wayland | `grim` + `slurp` | X11 tools often fail or behave oddly on Wayland | | Linux X11 / headless CI | `scrot` or browser-native capture | Works in minimal or virtual-display environments | | Windows desktop capture | `nircmd savescreenshot` or Pillow `ImageGrab` | Easier than verbose PowerShell screen APIs | | Web page or web app | Playwright | Best for stable viewport, element, full-page, masked, and regression screenshots | | Visual diff / screenshot tests | Playwright with fixed viewport | Better control over animations, caret, masks, and reproducibility | Default to the most native capture path first. Move to browser-native tooling when determinism, masking, element capture, or visual regression matters more than convenience. ## Core Rules ### 1. Pick the capture path by artifact, not by habit - Desktop UI screenshots usually want OS-native tools. - Web pages and web apps usually want browser-native capture, not a desktop screenshot of the browser window. - Simulator screenshots should come from the simulator tooling when possible. - Use region, window, or element capture when the point is local; use full screen or full page only when the full context matters. ### 2. Stabilize the target before capturing - Dynamic pages should settle before capture: wait for network idle or the specific element that matters, then give fonts and transitions a brief moment to finish. - Do not take the screenshot before the real rendered state exists. - For browser capture, prefer explicit readiness over blind sleeps when possible. - If the page never truly goes idle, wait for the exact UI state you need instead of chasing perfect stillness. ### 3. Freeze viewport, scale, zoom, and theme for reproducibility - Screenshot comparisons are meaningless if viewport, zoom level, theme, or device scale changed. - For browser captures, fix the viewport before taking baselines or before/after images. - Retina and HiDPI displays can produce more pixels than expected; decide whether you want physical pixels or CSS-scale output and keep that choice consistent. - If dark/light mode matters, capture both intentionally instead of mixing them accidentally. ### 4. Capture the smallest useful scope - Element, region, or window screenshots are usually better than noisy full-screen captures. - Full-page screenshots are useful for audits and archives, but long pages become hard to read and compare. - For browser work, element screenshots or clipped regions usually produce cleaner diffs than full-page output. - If the screenshot is evidence, keep enough surrounding context that the user can understand what they are looking at. ### 5. Remove noise before you capture - Hide or avoid unstable UI when it is not the subject: cursors, carets, toasts, chat widgets, notifications, loading spinners, timestamps, and randomized content. - Mask or avoid secrets, personal data, tokens, and internal URLs before capture. - For Playwright-style browser capture, features like disabled animations, hidden carets, and masking are worth using when visual stability matters. - If the noise is the bug, keep it; otherwise remove it. ### 6. Use the right output format - PNG is the default for screenshots, UI, code, terminals, and text-heavy captures. - JPEG is for photographic content, not normal screenshots. - WebP is fine for sharing or storage when compatibility is acceptable, but do not default to it if the consumer expects plain PNG files. - Avoid recompressing screenshots through JPEG pipelines unless the user explicitly wants smaller lossy output. ### 7. Make automation and CI captures debuggable - On failures, save a screenshot immediately before retrying or moving on. - Use stable filenames for baselines and timestamps for ad hoc or batch captures. - In CI, identical viewport and deterministic state matter more than raw screenshot volume. - Headless runs should prefer browser-native screenshots over trying to screen-grab the host display. ### 8. Validate that the screenshot is actually useful - Check that the important detail is visible, legible, and not cropped away. - Verify that secrets are not still visible in tabs, sidebars, URLs, notifications, or test data. - Before/after comparisons should use the same viewport, zoom, theme, and state. - A screenshot is bad if it is technically correct but useless for the human who needs it. ## High-Value Patterns - macOS: `screencapture -x out.png` for silent capture, `-i` for interactive selection, `-R x,y,w,h` for a fixed region. - iOS Simulator: `xcrun simctl io booted screenshot out.png` - Linux Wayland: `grim -g "$(slurp)" out.png` - Playwright page capture: wait for the target state, then use page, element, clipped, or full-page screenshots deliberately. - Playwright stability features worth remembering: fixed viewport, disabled animations, hidden caret, masks for sensitive regions, and stable theme/media settings. ## Common Traps - Taking a browser-window screenshot when an element or page screenshot was the real need. - Capturing before fonts, data, or layout transitions finish. - Comparing screenshots with different viewport sizes or zoom levels and treating the diff as meaningful. - Using JPEG for screenshots and blurring text, edges, and code. - Letting timestamps, cursor blinks, notifications, or random data ruin visual diffs. - Forgetting that Wayland breaks familiar X11 screenshot tools. - Sharing screenshots with secrets still visible in tabs, sidebars, URLs, or test accounts. - Taking full-page captures of huge pages and ending up with unreadable evidence. ## Related Skills Install with `clawhub install ` if user confirms: - `playwright` — Browser automation, DOM interaction, and web screenshots - `image` — Post-capture format, cropping, compression, and export decisions - `image-edit` — Annotation, cleanup, masking, and targeted edits after capture - `documentation` — Turning screenshots into docs, guides, and release assets - `video` — When a flow should be recorded instead of reduced to still images ## Feedback - If useful: `clawhub star screenshot` - Stay updated: `clawhub sync`

相关技能

其他

Self Improving Agent

掌握学习、错误和校正,以便不断改进。 使用时间:(1) 命令或操作意外失败, (2) 用户更正 Claude ('No, that's wrong...', 'Actually...'), (3)用户请求具备不存在的能力,(4)外部API或工具失败,(5)克洛德意识到其知识已经过…

其他

Self Improving + Proactive Agent

自我反省+自我批评+自我学习+自我组织记忆. 特工评估自己的工作,发现错误,并永久改进. 当 (1) 命令,工具, API, 或操作失败时使用; (2) 用户纠正或拒绝您的工作; (3) 您意识到您的知识已经过时或不正确; (4) 您发现更好的方法; (5) 用户明确安装或引用当…

其他

Proactive Agent

将AI代理人从任务跟踪者转变为预先预测需要并不断改进的主动伙伴. 现在WAL协议,工作缓冲,自主克龙, 和战斗测试模式。 哈尔·斯塔克的一部分

其他

Multi Search Engine

多搜索引擎集成16个引擎(7个CN+9 Global). 支持高级搜索操作员,时间过滤器,站点搜索,隐私引擎,和WolframAlpha知识查询. 不需要 API 密钥 .